The early fault signal characteristics of the rolling bearing are weak. The performance of traditional variational mode decomposition （VMD） depends on the parameters， which include mode number and penalty coefficient. To solve this problem， an adaptive method to determine parameters of VMD was proposed. The minimum center frequency quotient algorithm was proposed based on the idea of the center frequency to judge whether the bend limited intrinsic mode function （BIMF） is overlapped or not，and the mode number was determined by the ratio of the minimum frequency and the sub small frequency. By using the fuzzy entropy principle， the sum fuzzy entropy （SFE） was proposed to optimize the penalty coefficient. The correlation coefficient was used to select the BIMFs. The method can adaptively determine the mode number and penalty factor， suppress the mode aliasing phenomenon. it can extract the fault signal features from the strong noise to judge the bearing state.